transistors Department of Electrical and Computer Engineering By Yaman Sangar Amitesh Narayan Snehal Mhatre Overview Motivation Introduction CMOS vs CNTFETs CNT Technology Challenges Probabilistic model of ID: 776479
Download Presentation The PPT/PDF document " Probabilistic modelling of performance ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Probabilistic modelling of performance parameters of Carbon Nanotube transistors
Department of Electrical and Computer Engineering
By
Yaman Sangar
Amitesh Narayan
Snehal Mhatre
Slide2Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology - ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
1
Slide3Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
2
Slide4MOTIVATION: Why CNTFET?
04/29/2014
3
Dennard Scaling might not last long
Increased performance by better algorithms?
More parallelism?
Alternatives to CMOS -
FinFETs
,
Ge
-nanowire FET, Si-nanowire FET, wrap-around gate
MOS,
graphene
ribbon FET
What about an inherently faster and less power consuming device?
Yay CNTFET – faster with low power
Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
4
Slide65
Carbon Nanotubes
04/29/2014
Slide76
Types of CNTs
04/29/2014
Single Walled CNT (SWNT)Double Walled CNT (DWNT)Multiple Walled CNT (MWNT)Depending on Chiral angle:Semiconducting CNT (s-CNT)Metallic CNT (m-CNT)
Slide87
Properties of CNTs
04/29/2014
Strong and very flexible molecular material
Electrical conductivity is 6 times that of copper
High current carrying capacity
Thermal conductivity is 15 times more than copper
Toxicity?
04/29/2014
8
CNTFET
How CNTs conduct?
Gate used to electrostatically induce carriers into tubeBallistic Transport
Slide10Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
9
Slide11CircuitFETDelay (In Picoseconds)Power (In uWatts)InverterCMOS16.589.81CNT3.780.252 Input NandCMOS24.3220.67CNT5.980.692 Input NorCMOS39.2622.13CNT6.490.48
Simulation based Comparison
between CMOS and CNT technology
04/29/2014
10
Slide1204/29/2014
11
Better delay
Circuit
FETDelay (In Picoseconds)Power (In uWatts)InverterCMOS16.589.81CNT3.780.252 Input NandCMOS24.3220.67CNT5.980.692 Input NorCMOS39.2622.13CNT6.490.48
Simulation based Comparison
between CMOS and CNT technology
Slide1304/29/2014
12
Better delay
At lower power!
CircuitFETDelay (In Picoseconds)Power (In uWatts)InverterCMOS16.589.81CNT3.780.252 Input NandCMOS24.3220.67CNT5.980.692 Input NorCMOS39.2622.13CNT6.490.48
Simulation based Comparison
between CMOS and CNT technology
Slide14Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
13
Slide15Major CNT specific variationsCNT density variationMetallic CNT induced count variationCNT diameter variationCNT misalignmentCNT doping variation
Challenges with CNT technology
04/29/2014
14
Unavoidable process variations
Performance parameters affected
Slide16CNT density variation
CNT diameter variation
Current variation
Threshold voltage variation
04/29/2014
15
Slide17CNT Misalignment
CNT doping variation
Changes effective CNT length
Short between CNTs
Incorrect logic functionality
Reduction in drive current
May not lead to unipolar behavior
04/29/2014
16
Slide18Metallic CNT induced count variation
m-CNT
s-CNT
Excessive leakage current
Increases power consumption
Changes gate delay
Inferior noise performance
Defective functionality
s
-CNT
m
-CNT
Vgs
Current
04/29/2014
17
Slide1918
Removal of m-CNTFETs
VMR Technique : A special layout called VMR structure consisting of inter-digitated electrodes at minimum metal pitch is fabricated. M-CNT electrical breakdown performed by applying high voltage all at once using VMR. M-CNTs are burnt out and unwanted sections of VMR are later removed.Using Thermal and Fluidic Process: Preferential thermal desorption of the alkyls from the semiconducting nanotubes and further dissolution of m-CNTs in chloroform.Chemical Etching: Diameter dependent etching technique which removes all m-CNTs below a cutoff diameter.
04/29/2014
Slide20Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
19
Slide21Probabilistic model of CNT count variation due to m-CNTs
P N gs = n gs | N = n = n C n gs p s n gs p m ( n − n gs ) P N gm = n gm | N = n = n C n gm p s ( n − n gm ) p m n gm
Slide22Conditional probability after removal techniques
Ns = number of surviving s-CNTsNm = number of surving m-CNTsprs = conditional probability that a CNT is removed given that it is s-CNT prm = conditional probability that a CNT is removed given that it is m-CNTqrs = 1 - prsqrm = 1 -prm
04/29/2014
21
Slide23Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
22
Slide24Effect of CNT count variation on ION / IOFF tuning ratio
Slide25Current of a single CNT
ICNT = ps Is + pmImµ(ICNT) = psµ( Is )+ pmµ(Im )ICNT = drive current of single CNT (type unknown)Is = drive current of single s-CNTIm = drive current of single m-CNTps = probability of s-CNTpm = probability of m-CNT
04/29/2014
24
Slide26Ns = count of s-CNTNm = count of m-CNTIs,on = s-CNT current, Vgs = Vds = VddIs,off = s-CNT current, Vgs = 0 and Vds = VddIm = m-CNT current, Vds = Vdd
ION / IOFF ratio of CNTFET
04/29/2014
25
Slide27µ( 𝐼 𝑂𝑁 ) µ ( 𝐼 𝑂𝐹𝐹 ) = 𝑝 𝑠 1 − 𝑝 𝑟𝑠 𝜇 ( 𝐼 𝑠 , 𝑜𝑛 ) + 𝑝 𝑚 1− 𝑝 𝑟𝑚 𝜇 ( 𝐼 𝑚 ) 𝑝 𝑠 1 − 𝑝 𝑟𝑠 𝜇 ( 𝐼 𝑠 , 𝑜𝑓𝑓 ) + 𝑝 𝑚 1− 𝑝 𝑟𝑚 𝜇 ( 𝐼 𝑚 ) µ( 𝐼 𝑂𝑁 ) µ ( 𝐼 𝑂𝐹𝐹 ) = µ( 𝑁 𝑠 ) µ( 𝐼 𝑠 , 𝑜𝑛 ) + µ( 𝑁 𝑚 ) µ( 𝐼 𝑚 ) µ( 𝑁 𝑠 ) µ( 𝐼 𝑠 , 𝑜𝑓𝑓 ) + µ( 𝑁 𝑚 ) µ( 𝐼 𝑚 )
Slide28Effect of various processing parameters on the ratio µ(ION) / µ(IOFF)
µ(ION) / µ(IOFF) is more sensitive to prmµ(ION) / µ(IOFF) = 104 for prm > 1 – 10 -4 = 99.99 % for pm = 33.33%
04/29/2014
27
1-
prm
Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
28
Slide3029
04/29/2014
Effect of CNT count variation on Gate delay
30
04/29/2014
= =
31
Plot of v/s
04/29/2014
= 0.3
N = 10
N = 20
N = 30
N = 40
N = 50
Slide3332
04/29/2014
Plot of v/s N
N
0.2
0.4
0.6
0.8
0.9
Slide34Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
33
Slide3534
Noise Margin of CNTFET
04/29/2014
Slide3635
VIL and VIH
Substituting = Vin, , and = Differentiating with respect to Vin and substituting -1
04/29/2014
nFET
p
FET
Slide3704/29/2014
36
For CNTFET,
For CMOS,
V
IL
and V
IH
NM
L
= V
IL
- 0
NM
H
= V
DD
– V
IH
Slide38Overview
MotivationIntroductionCMOS v/s CNTFETsCNT Technology – ChallengesProbabilistic model of faultsModelling performance parameters:ION / IOFF tuning ratioGate delayNoise MarginConclusion
04/29/2014
37
Slide3938
CONCLUSION
Modeled count variations and hence device current as a probabilistic functionStudied the affect of these faults on tuning ratio and gate delayInferred some design guidelines that could be used to judge the correctness of a processMathematically derived noise margin based on current equations – better noise margin than a CMOS
04/29/2014
Slide40